1 2 3 47 6 Journal of Integer Sequences, Vol. 17 (2014), Article 14.1.8 23 11 On a Conjecture of Farhi Soufiane Mezroui, Abdelmalek Azizi, and M’hammed Ziane Laboratoire ACSA Département de Mathématiques et Informatique Université Mohammed Premier Oujda 60000 Morocco mezroui.soufiane@yahoo.fr abdelmalekazizi@yahoo.fr ziane12001@yahoo.fr Abstract Recently, Farhi showed that every natural N 6≡ 2 (mod 24) can be written j 2number k n as the sum of three numbers of the form 3 (n ∈ N). He conjectured that this result remains true even if N ≡ 2 (mod 24). In this note, we prove this statement. 1 Introduction Throughout this note, we let N and Z, respectively, denote the set of the non-negative integers and the set of the integers. We let ⌊·⌋ and h·i denote the integer-part and the fractional-part functions. Let X be a set. We denote the cardinality of X by #X. We also recall that ·· is the Jacobi symbol. Recently, Farhi [1] showed that every natural j 2 k number N 6≡ 2 (mod 24) can be written as the sum of three numbers of the form n3 (n ∈ N). He conjectured that this result remains true even if N ≡ 2 (mod 24). We recall his conjecture. Conjecture j 2 k 1. Every natural number can be written as the sum of three numbers of the form n3 (n ∈ N). In fact, he proposed a more general conjecture. 1 Conjecture 2. Let k ≥ 2 be an integer. There then exists a positive integer a(k) that satisfies the followingjproperty: every natural number can be written as the sum of k + 1 k nk numbers of the form a(k) (n ∈ N). In this note, we prove Conjecture 1. 2 Proof of Conjecture 1 We recall Legendre’s theorem [3, pp. 331–339], which is a necessary tool for our proof: Theorem 3. Every natural number not of the form 4h (8k + 7)(h, k ∈ N) can be represented as the sum of three squares of natural numbers. We note that since 4h (8k + 7) is congruent to 0, 4 or 7 modulo 8, every natural number not congruent to 0, 4 or 7 modulo 8 can be represented as the sum of three squares of natural numbers. We will use this result later. Let r3 (n) be the number of representations of the positive integer n as the sum of three squares of integers. The following theorem provides an interesting formula for r3 (n), which can be proven using the theory of modular functions. Theorem 4 (see [2]). For any positive integer n, we have Y 1 1 1 16 √ nχ2 (n)K(−4n) 1 + + · · · + b−1 + b r3 (n) = π p p p 2 p |n 1− −p−2b n p −1 ! 1 , p where b = b(p) is the largest integer such that p2b | n, ∞ X −4n 1 , K(−4n) = m m m=1 and if 4a is the highest power of 4 dividing n, then if 4−a n ≡ 7 (mod 8); 0, χ2 (n) = 21a , if 4−a n ≡ 3 (mod 8); 3 , if 4−a n ≡ 1, 2, 5, 6 (mod 8). 2a+1 We will require the following technical lemma. Lemma 5. For any positive integer n ≡ 1 (mod 8), we have r3 (9n) > 2 3 r3 (n). 2 ✷ Proof. We have r3 (9n) = Y 1+ p2 |9n ′ 1 1 + ··· + ′ b p p −1 16 √ 9nχ2 (9n)K(−36n)× π ! !−1 ′ −2b −9 p n 1 1 , + ′ 1− b p p p ′ ′ where b = b (p) denotes the largest integer for which p2b | 9n. Since n ≡ 1 (mod 8), it follows that 40 = 1 is the highest power of 4 dividing n. This result implies that χ2 (n) = 23 . Similarly, we have 9n ≡ 1 (mod 8). Thus, 40 = 1 is the highest power of 4 dividing 9n, which gives χ2 (9n) = χ2 (n) = 23 . Conversely, it follows from [2, p. 84] that −4n 1 2 K(−36n) = K(−4 × 3 × n) = 1 − K(−4n). 3 3 Since n ≡ 1 (mod 8), it follows from Legendre’s theorem that n can be represented as the sum of three squares of natural numbers. Thus, r3 (n) 6= 0. Dividing through by r3 (n) then yields an identity equivalent to −1 ! ′ Q −2b 1 + p1 + · · · + b′1−1 + 1b′ 1 − −9 pp n p1 p2 |9n p p 3 r3 (9n) × = −2b −1 · 1 −1 Q r3 (n) −p n 1 1 − −4n 1 1 1 3 3 p2 |n 1 + p + · · · + pb−1 + pb 1 − p p ′ ′ ′ Let p 6= 3 with p2 | n. Thus, b = b (p) is the largest integer for which p2b | n. Therefore, ′ ′ one obtains b = b (p) = b(p) = b. Furthermore, we have ! ! ! ′ ′ ′ −p−2b n 32 −p−2b n −p−2b n −9 p−2b n = = = · p p p p p −1 ′ −9 p−2b n 1 1 1 1 2 For every p 6= 3 with p | n, we then have 1 + p + · · · + b′ −1 + b′ 1 − = p p p p −2b −1 1 . Thus, two cases are evident: if 32 | n, then + p1b 1 − −p p n p1 1 + p1 + · · · + pb−1 r3 (9n) = r3 (n) 1− 3 −4n 3 1 3 1 −1 × 3 1 + + ··· + 1 3b ′ 1 + 31 + · · · + −1 + 1 3b−1 Otherwise, 32 does not divide n, so 1 3 1 r3 (9n) = 1 −1 × 1 + · · · + b′ −1 + b′ −4n r3 (n) 3 3 1− 3 3 3 1 ′ 3b + 1 3b ′ −9×3−2b n 3 −1 1 1− 3 −1 −2b 1 − −3 3 n 31 1− −9 × 3 3 −2b ′ , ! !−1 n 1 . 3 We now show that in all cases, r3 (9n) > ′ 3 2 r3 (n). ′ • If 32 does not divide n, b = b (3) = 1 is implied to be the largest integer for which ′ 32b | 9n. One obtains −1 ! 1 −n 1 3 r3 (9n) . 1− = 1 −1 × 1 + r3 (n) 3 3 3 1 − −4n 3 3 3 1 We have 1 − −4n > 23 , which gives the result = 1, 23 or 34 and so 1 −1 3 3 (1−( −4n 3 )3) r3 (9n) > 32 r3 (n). ′ • If 32 | n, then b (respectively b ) is the largest integer for which 32b | n (respectively ′ 32b | 9n). Hence, −1 1 1 −9×3−2(b+1) n 1 1 1 + 3 + · · · + 3b + 3b+1 1 − 3 3 r3 (9n) 3 × = −1 −1 1 −4n −2b r3 (n) 1 1− 3 3 + 31b 1 − −3 3 n 13 1 + 13 + · · · + 3b−1 −2b −1 1 −3 n 1 1 1 + · · · + 1 − + 1 + 3 3 3 3b 3b+1 3 = × −1 · −1 1 1 1 −3−2b n 1 1 1 − −4n 1 + 3 + · · · + 3b−1 + 3b 1 − 3 3 3 3 −2b We have 1 − −3 3 n 13 = 1, 32 or 43 . One obtains the following in all cases: 1 1 + b+1 b 3 3 1− This result implies 1 + 31 + · · · + 31b −2b −1 1 1 − −3 3 n 31 . Conversely, 3b result, r3 (9n) > 3 2 −1 −2b −1 1 1 −3 n 1 ≥ b 1− . 3 3 3 3 −2b −1 1 ≥ 1 + 31 + · · · + + 3b+1 1 − −3 3 n 13 −3−2b n 3 3 −4n 1 −1 1− ( ( 3 )3) > 3 . 2 1 3b−1 + Thus, we obtain the desired r3 (n). Theorem 6. Every jnatural k number N ≡ 2 (mod 24) can be written as the sum of three n2 numbers of the form 3 (n ∈ N). Proof. We may write N = 2 + 24k with k ∈ N. Thus, 3N + 3 = 9(1 + 8k). We now define two sets S1 and S2 as follows: n o 3 2 2 2 S1 = (a, b, c) ∈ Z : a + b + c = 1 + 8k , n o S2 = (a, b, c) ∈ Z3 : a2 + b2 + c2 = 9(1 + 8k) . 4 By the definition of r3 , we have #S2 = r3 (9(1 + 8k)) and #S1 = r3 (1 + 8k). Since 1 + 8k ≡ 1 (mod 8), we apply Lemma 5 to obtain r3 (9(1 + 8k)) > 23 r3 (1 + 8k) ≥ r3 (1 + 8k). One obtains r3 (9(1 + 8k)) > r3 (1 + 8k), which is equivalent to #S2 > #S1 . We note that this last result is the key to the proof. Let us define the map f : S1 −→ S2 (a, b, c) 7−→ (3a, 3b, 3c). We see easily that f is well defined and injective. Since #S2 > #S1 , we can find (a, b, c) ∈ S2 such that (a, b, c) ∈ / f (S1 ). Furthermore, we have a2 + b2 + c2 = 9(1 + 8k) ≡ 0 (mod 3), 2 2 then either a ≡ b ≡ c2 ≡ 1 (mod 3) or a2 ≡ b2 ≡ c2 ≡ 0 (mod 3). The last case cannot hold because one of the elements, a, b and c, is not divisible by 3 ((a, b, c) ∈ / f (S1 )). Thus, a2 ≡ b2 ≡ c2 ≡ 1 (mod 3) and we have N + 1 = 3(1 + 8k) a2 b 2 c 2 = + + 3 2 3 23 2 2 2 2 b c a b c a + + + + + . = 3 3 3 3 3 3 2 2 2 1 1 1 a b c 2 2 2 + + = + + = 1, which gives Since a ≡ b ≡ c ≡ 1 (mod 3), then 3 3 3 3 3 3 2 2 2 a b c N = + + . We replace (a, b, c) ∈ Z3 by (|a|, |b|, |c|) ∈ N3 to obtain the 3 3 3 desired solution. The conjecture is proven. 3 Acknowledgments The authors would like to thank the editor and the anonymous reader for their invaluable comments. This work was supported by Académie Hassan 2 under the project “Mathématiques et Applications: cryptographie” and URAC6-CNRST. References [1] B. Farhi, On the representation of the natural numbers as the sum of three terms of the 2 sequence ⌊ na ⌋, J. Integer Seq. 16 (2013), Article 13.6.4. [2] P. T. Bateman, On the representation of a number as the sum of three squares, Trans. Amer. Math. Soc. 71 (1951), 70–101. 5 [3] A. M. Legendre, Théorie des Nombres, 3rd ed., Vol. 2, 1830. 2010 Mathematics Subject Classification: Primary 11B13. Keywords: additive base, Legendre’s theorem, representation of an integer as the sum of three squares. Received July 25 2013; revised versions received July 31 2013; November 22 2013; December 30 2013. Published in Journal of Integer Sequences, December 30 2013. Revision, January 3 2014. Return to Journal of Integer Sequences home page. 6